Buried Underwater Target Classification using the New BOSS and Canonical Correlation Decomposition Feature Extraction

Abstract

Multi-aspect detection and classification of buried underwater objects using the new Buried Object Scanning Sonar (BOSS) data is the main goal of this project. Canonical coordinate decomposition (CCD) was applied to extract the most coherent features of the buried or bottom objects in two sonar pings with a certain separation. CCD provides an elegant framework for analyzing linear dependence and mutual information between two data channels. These features are then used for subsequent classification. For this study, single-aspect and multi-aspect classification schemes are evaluated, and the results presented in terms of confusion matrices. Additionally, the results of applying both the single and multi-aspect classifiers to the entire test runs are presented to show the real usefulness of the algorithms for buried/bottom mine-hunting.

Undersea Target Classification Using Canonical Correlation Analysis

“…Moreover, we showed that in a fixed bottom condition, canonical correlation features are relatively invariant to changes in aspect angle. Recent results [23] on buried object scanning sonar (BOSS) data collected in St. Andrew's Bay, Panama City, FL, further validate the potential of the proposed canonical-correlation-based feature extraction method for detection and classification of buried underwater objects.…”

Buried Underwater Object Classification Using a Collaborative Multiaspect Classifier

“…This system is selected since the previous studies [3], [4] indicated its excellent results in buried underwater target classification. In this system, a two-layer BPNN with 40 neurons in the hidden layer and two neurons in the output layer is used to provide intermediate decisions regarding the class of an object based upon the same CCA features that are used with the CMAC system.…”

“…When used to perform simultaneous detection and classification on data from an entire run, it was shown that the CMAC system was able to correctly detect and classify all of the mine-like objects and substantially reduce the number of false alarms not removed by the nonlinear DLF system. The adaptability and ease of implementation of this system coupled with its superior performance on all the BOSS tested data sets [3], [22], [23] makes it a valuable tool for underwater buried object classification.…”

“…One way to accomplish this is to use canonical correlation analysis (CCA) [1], which allows coherence-based features to be extracted from pairs of sonar pings. Using these coherence-based features allows for successful discrimination between mine-like and nonmine-like objects [2], [3].…”

Classification of Objects in Synthetic Aperture Sonar Images

“…There are a wide variety of techniques that are being developed to actually detect objects of interest. Many techniques rely on the signal strength, shape information of the imaged objects, or the amount of linear dependence (coherence) between two sonar returns of different grazing angles for the classification of objects [3,5]. However there are several problems with using such information for classification.…”